machine learning for risk management in matlab · portfolios and project returns using monte carlo...
Post on 06-Apr-2020
14 Views
Preview:
TRANSCRIPT
1© 2013 The MathWorks, Inc.
Machine Learning for Risk Management in MATLAB
Marshall AlphonsoSenior Engineer - FinanceMathWorksMarshall.Alphonso@mathworks.com
2
Agenda
What is machine learning?
Machine learning in MATLAB– Bond classification– Credit classification– Trading strategy
Moving models to production
3
What is machine learning?“[Machine Learning] gives computers the ability to learn without being explicitly programmed” Arthur Samuel, 1959
5
Deep learning is available too!
Deep refers to the number of hidden layers
Neural Network2-3 Hidden Layers
Deep Learning ModelsAs many as 150!
6
Machine learning applications in finance
F(x)Input/
Predictors Output/
Response
Energy Forecasting
,...),,( DPtTfEL =
Trading Rating score card
7
Machine learning challengesData challenges Volume of data is growing Velocity of data is accelerating Variety of data is dynamic Data cleaning is time consuming
Modeling challenges Data driven models No “one size fits” all solution Machine learning modeling is iterative
Production challenges Scalability – leveraging IT resources Flexibility – interfacing with systems
ManagementTraders
Quant Group
Financial Engineer
Regulators Clients Partners
Other groups
8
Demo: Calibrating the Rating System
Overseeing a portfolio of bonds Improve rating engine using machine learning
X y
9
Demo: Trading strategy
Hand Written Program Formula or Equation
If RSI > 70then “SELL”
If MACD > SIG and RSI <= 70then “HOLD”
…
𝑌𝑌Trade= 𝛽𝛽1𝑋𝑋RSI + 𝛽𝛽2𝑋𝑋𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+ 𝛽𝛽3XTSMom +
…
“[Machine Learning] gives computers the ability to learn without being explicitly programmed” Arthur Samuel, 1959
ComputerProgram
Buy
Sell
Hold
Standard Approach
Prediction = F(factors, trade decision)
Inputs → Outputs
Machine Learning
Buy
Sell
Hold
Machine Learning Approach
10
Modeling is an art formModeling in MATLAB is visual and requires less code
• APPs facilitate analysis, but with auto codegen transparency
• Lots of high quality financial & machine learning functionality out of the box!
• You don’t have to be a programmer to do analysis
11
Agenda
What is machine learning?
Machine learning in MATLAB– Bond classification– Trading strategy
Moving models to production
12
MATLAB Programs Can be Shared With Anyone
Share With Other MATLAB Users Share With People Who do Not Have MATLAB
14
MATLAB is flexible
MATLAB
C/C++ExcelAdd-in JavaHadoop .NET
MATLABCompiler
MATLABProduction
Server
StandaloneApplication
MATLABCompiler SDK
Apps Files
Custom Toolbox
Python
With MATLAB Users
With People Who Do Not Have MATLAB
15
MATLAB integrates with IT systems
Databases
CloudStorage
IoT &Big Data
Visualization
Web
Custom App
Public Cloud Private CloudPlatform
Data Business System
AzureBlob
AzureSQL Request
Broker
MATLAB Production Server
MDCS
AzureIoT Hub
MATLAB
16
Machine learning challengesData challenges Volume of data is growing Velocity of data is accelerating Variety of data is dynamic Data cleaning is time consuming
Modeling challenges Data driven models No “one size fits” all solution Machine learning modeling is iterative
Production challenges Scalability – leveraging IT resources Flexibility – interfacing with systems
ManagementTraders
Quant Group
Financial Engineer
Regulators Clients Partners
Other groups
17
Why MATLAB for Machine Learning?Data challenges MATLAB works with BIG data Lot of preprocessing functions Point & click speeds up exploration
Modeling challenges APPs simplify modeling Iterate quickly with parallel Interactive visuals generate insight
Production challenges Scalability – leverages cloud Flexibility – interfaces IT systems
ManagementTraders
Quant Group
Financial Engineer
Regulators Clients Partners
Other groups
18
Financial
Statistics & Machine Learning Optimization
Financial Instruments
Econometrics
Risk Management
Machine learning workflow
MATLAB
Parallel Computing
MATLAB Distributed Computing Server
MATLAB Compiler SDK
MATLAB Compiler
Report G
enerator
Production Server
Datafeed
Database
Spreadsheet Link
Trading
Neural Networks
Curve Fitting
Symbolic Math
Signal Processing
Global Optimization
Files
Databases
Datafeeds
AccessData Analysis and Visualization
Financial Modeling
Application Development
Research and QuantifyReporting
Applications
Production
Share
19
Want to learn more?
Marshall AlphonsoSenior Engineer - FinanceMarshall.Alphonso@mathworks.com
Mike DeLuciaAccount ManagerMike.DeLucia@mathworks.com
Chuck CastriconeAccount ManagerChuck.Castricone@mathworks.com
Account Managers
New York Engineer
Contact us!
20© 2013 The MathWorks, Inc.
© 2017 The MathWorks, Inc. MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand names may be trademarks or registered trademarks of their respective holders.
22
MathWorks at a Glance
● Office locations ● Distributors serving 16 countries
Earth’s topography on a Miller cylindrical projection, created with MATLAB and Mapping Toolbox.
Revenues ~$850M in 2016 More than 60% from outside the U.S.
Privately held 3500 employees worldwide More than 2 million users in 175+ countries
23
Key Industries
Aerospace and defense
Automotive Biological sciences Biotech and
pharmaceutical Communications Electronics Energy production Financial services
Industrial automation and machinery
Medical devices Metals, materials, and
mining Neuroscience Railway systems Semiconductors Software and internet
24
Deeply Rooted in Education
Benefits for Industry Every year, tens of thousands of engineers enter the workforce with
MATLAB and Simulink product skills and experience. Students learn theory and techniques while using MATLAB and Simulink.
5000+ universities around the world 1800+ MATLAB and Simulink based books Academic support for research, fellowships, student
competitions, and curriculum development
“Everyone that comes in as a new hire already knows MATLAB, because they all had it in college. The learning curve is significantly lessened as a result.”Jeff Corn, Chief of Engineering Projects Section,U.S. Air Force
25
Core MathWorks Products
The leading environment for technical computing
The industry-standard, high-level programming language for algorithm development
Numeric computation Parallel computing, with multicore and multiprocessor
support Data analysis and visualization Toolboxes for signal and image processing, statistics,
optimization, symbolic math, and other areas Tools for application development and deployment Foundation of MathWorks products
26
Training ServicesExploit the full potential of MathWorks products
Flexible delivery options: Public training available worldwide Onsite training with standard or customized courses Web-based training with live, interactive instructor-led
courses Self-paced interactive online training
More than 50 course offerings: Introductory and intermediate training on MATLAB, Simulink,
Stateflow, code generation, and Polyspace products Specialized courses in machine learning, control design,
signal processing, parallel computing, code generation, communications, financial analysis, and other areas
27
Technical SupportResources Over 100 support engineers
– Most with MS degrees (EE, ME, CS)
– Local support in North America, Europe, and Asia
Comprehensive, product-specific web support resources
High customer satisfaction 95% of calls answered
within three minutes 70% of issues resolved
within 24 hours 90% of customers surveyed said they
were satisfied or very satisfied
28
The MathWorks Advantage
Mission to develop leading-edge tools Leading environments for technical computing and
Model-Based Design– Unified, integrated computing environment– Modeling, simulation, and prototyping, combined with broad
analysis capabilities– Implementation to desktop environments and embedded
systems– Integrated design and test
Open systems philosophy and architecture Multiplatform support and interoperability
29
The MathWorks Advantage
Strong, customer-focused company – Heavy investments in R&D– Partnership with customers to help define product capabilities
Breadth of supported applications Consistent, highly rated service and support Prominence of MATLAB in engineering education Partnerships with industry leaders
31
CAMRADATA Models Dependencies for Quantitative Risk Assessment with MathWorks Tools
ChallengeRapidly develop quantitative tools for factor analysis,risk analysis, and defensive asset allocation
SolutionUse MATLAB to model complex non-linear dependencies between assets, liabilities, and economic variables using copulas
Results Development time reduced by 90 percent Risk calculated in hours, not weeks Diverse skill sets leveraged
“Using MATLAB we can build a model in one morning. It would take two weeks to write the equivalent code in Visual Basic.”
Martyn DoreyCAMRADATA
Risk-assessment model developed in MATLAB.
Link to user story
32
Capgemini Helps Clients Achieve Basel IICompliance and Deliver Economic Capital,Risk, and Valuation Models with MATLAB
ChallengeEnable banking clients to meet Basel II regulatory guidelines and perform other risk management tasks
SolutionUse MATLAB to develop risk management models and to perform valuations of complex products
Results Strong competitive advantage established Scalable solution delivered Customer portfolio revalued
“With its computational power, matrix infrastructure, and ability to perform Monte Carlo simulations, MATLAB gives us a competitive advantage in performing complex risk analyses."
Dr. Marco FolpmersCapgemini
Link to user story
Scatterplots showing 500,000 simulations drawn from bivariate t-copulas with the same correlation coefficient but differing degrees of freedom.
33
Intuitive Analytics Uses MATLAB to Build Quantitative Tools to Help Bond Issuers Manage Risk
ChallengeBuild and market a quantitative tool for reducing expected cost and risk for municipal bond issuers
SolutionUse MathWorks tools to develop algorithms, visualize results, and simplify deployment of an advanced analytical tool
Results Development productivity increased by 90% Deployment simplified Visual environment created
“Because MATLAB enables us to build and distribute applications to analysts that are accessible from Excel, we are quickly bringing to market products that are adopted and deployed by investment banks.”
Peter OrrIntuitive Analytics
Using MATLAB technical computing software to provide visual representations of interest rate models.
Link to user story
34
IPD Develops and Deploys Real Estate Cash Flow Models with MathWorks Tools
ChallengeCreate cash flow models of real estate investment portfolios and project returns using Monte Carlo simulations
SolutionUse MATLAB and MATLAB Builder NE to develop optimization algorithms, build financial models, and deploy solutions
Results Development time cut by 16 weeks Updates completed in hours Deployment simplified
“The only other approach we seriously considered involved developing a class library in .NET and C#. Development, debugging, and testing would have taken us 37 weeks. Using MATLAB, we completed the project in 21 weeks.”
Peter McAnenaInvestment Property Databank
MATLAB graph generated to indicate how total returns from industrial property are likely to behave.
'06
'07
'082 4 6 8 10 12 14 16 18
0%5%
10%15%20%25%
0-0.05 0.05-0.1 0.1-0.15 0.15-0.2 0.2-0.25
Link to user story
35
Nykredit Develops Risk Management and Portfolio Analysis Applications to Minimize Operational Risk
ChallengeEnable financial analysts to make rapid, fact-baseddecisions by providing them with direct access to riskmanagement and portfolio analysis information
SolutionDevelop and deploy easy-to-use graphical financialanalysis applications using MATLAB and MATLABCompiler
Results Productivity increased threefold Operational risk mitigated Analysis time reduced from days to hours
Link to user story
Nykredit’s tool for calculating and visualizing risk statistics. The plot shows portfolio expected tracking error broken out by industry.
“Data handling, programming, debugging, and plotting are much easier in MATLAB, where everything is in one environment. For performance calculation GUIs, MATLAB provides a real error-checked application that makes cool customized plots for client reports. This has turned a several-hour task in a spreadsheet into a two-minute no-brainer.”
Peter AhlgrenNykredit Asset Management
36
Macroeconomic Modeling and Inflation Rate Forecasting at the Reserve Bank of New Zealand
ChallengeSupport New Zealand monetary policy with a theoretically well-founded model
SolutionUse MATLAB to analyze and forecast macroeconomic variables, and communicate results to stakeholders
Results Entire workflow completed in a single environment Code shared with other central banks and
financial institutions Technical rigor of macroeconomic forecasting
increased
“With all RBNZ models now implemented in MATLAB, the RBNZ has a common platform for evaluating the economy and making informed decisions.”
Jaromir BenesInternational Monetary Fund
Link to article
Sample fancharts produced by RBNZ’s macroeconomic model.
37
Robeco Develops Quantitative Stock Selection and Portfolio Optimization Models with MathWorks Tools
ChallengeDevelop, distribute, and maintain quantitative tools for portfolio construction and management
SolutionUse MATLAB and MATLAB Builder NE to develop algorithms, build quantitative models, and deploy solutions
Results Applications updated faster Black-box solutions eliminated Scalability and flexibility increased
“Unlike companies that rely on off-the-shelf quantitative analysis solutions, we can see our process improving all the time. We have the flexibility to continuously improve our algorithms and models in MATLAB—and that is a big advantage.”
Willem JellemaRobeco
Interest rate paths for the risk analysis of a savings product.
Link to user story
40
Machine learning is used everywhere!
•Computational finance, for credit scoring and algorithmic trading
•Image processing and computer vision, for face recognition, motion detection, and object detection
•Computational biology, for tumor detection, drug discovery, and DNA sequencing
•Energy production, for price and load forecasting
•Automotive, aerospace, and manufacturing, for predictive maintenance
•Natural language processing, for voice recognition applications
42
Write Your Programs OnceThen Share To Different Targets
MATLAB
C/C++ExcelAdd-in JavaHadoop .NET
MATLABCompiler
MATLABProduction
Server
StandaloneApplication
MATLABCompiler SDK
Apps Files
Custom Toolbox
Python
With MATLAB Users
With People Who Do Not Have MATLAB
43
Value of MATLAB Production Server
Directly deploy MATLAB programs into production– Supports multiple MATLAB programs and MCR versions
Scalable & reliable– Service large numbers of concurrent requests– Add capacity or redundancy with additional servers
Use with web, database & application servers– Lightweight client library isolates MATLAB processing
MATLAB Production Server(s)
HTMLXML
Java ScriptWeb
Server(s)
44
MODEL
PREDICTION
Predictive Modeling Workflow
Train: Iterate till you find the best model
Predict: Integrate trained models into applications
MODELSUPERVISEDLEARNING
CLASSIFICATION
REGRESSION
PREPROCESS DATA
SUMMARYSTATISTICS
PCAFILTERS
CLUSTER ANALYSIS
LOAD DATA
PREPROCESS DATA
SUMMARYSTATISTICS
PCAFILTERS
CLUSTER ANALYSIS
NEWDATA
45
What does Trading Toolbox do?
Trading Toolbox
Market Access
Financial Toolbox
Optimization Toolbox
Trading EngineFinancial Toolbox
Statistics Toolbox
Risk EngineFin. Instruments Tbx
Financial Toolbox
Pricing Engine
Bloomberg EMSXX_TraderInteractive BrokersCQG
Bloomberg EMSX
46
Create Credit Scorecards Determine Probability of Default Calculate Expected Loss
Demo: Lending club overview
EL LGDPD EAD
47
Create Credit Scorecards Determine Probability of Default Calculate Expected Loss
Demo: Consumer Credit Risk Modeling
EL LGDPD EAD
50
Demo: Volatility Modeling
2009 2010 2011 2012 20130.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08Volatility Model Comparison
Vol Historical
Vol GARCH 1 - GARCH with bias offset
Vol GARCH 2 - GARCH with no bias offset
Neural Net
51
Share applications with those who do not need MATLAB
Royalty free
MATLAB Production Server provides most efficient path for secure and scalable enterprise applications
MATLAB
MATLABCompiler SDK
C/C++ JavaHadoop .NET
MATLABCompiler
MATLABProduction
Server
Sharing MATLAB Applications
MATLAB .exe Excel
52
Machine learning workflow
Explore and Prototype
Data Analysis & Visualization
Financial Modeling
Application Development
Reporting
Applications
Production
Share
Scale
Files
Databases
Datafeeds
Access
Small/Big Data Predictive Modeling Deploy
top related